# Shaping the Future: Exploring the Impact of Generative AI on Youth
## (Talk Outline <Dan/Yim>)

### A "Teens-First" Approach to AI Literacy: Self-Advocacy Strategies for Teens to Navigate AI Impacts

Teens are concerned with all sorts of things in their daily lives -- life revolves around their grades, relationships, looks, health, getting into college, and maybe amping up their social media presence. For many of them, AI is not their central concern. Yet, AI is increasingly shaping all these areas, from software that grades assignments or writes essays to tools that design social media posts, create stunning art, or even enable harmful behaviors like personalized bullying campaigns (CITE). AI is both a powerful resource and a significant risk.

While engineering solutions like guardrails and parental controls aim to make AI interactions safer, research shows that restrictions can often backfire, prompting teens to find creative workarounds (CITE). This talk shifts the focus to a different approach: empowering teens as their own advocates. By equipping young people with the knowledge and skills to understand and navigate AI’s role in their lives, we can foster agency and critical thinking (CITE CITE). Drawing on research and practice, we’ll share strategies for centering teens in their own AI experiences through a blend of AI literacy and educational techniques that promote self-advocacy and meaningful engagement.

### Script

*Introductions* ... and today we are here to present a teens-first approach to AI literacy, through self-advocacy strategies for teens to navigate their own AI experiences and impacts of AI.

All of us here today are likely familiar with a bunch of different AI risks to teens and young people. We are concerned about bullying, disinformation, inaccurate health advice, cheating on tests and essays, and a whole bunch of other things that crop up what seems like every second.

But that view kind of looks at teens as inside this AI bubble – where AI is the center and it affects different populations. That’s true of course –

But we see it more like teens are this bubble, and AI is maybe a small part of that on their minds. We do know that Gen Z is interested in AI and even aware of the risks it poses, including climate impact. But overall, teens care about all sorts of things and AI is probably a very small part of that.

However, AI is actually impacting each of those areas of their life. AI can help write a better essay, get better grades. I can smooth my skin using AI beauty filters on social media. Kids can bully each other with AI generated images. AI can spread misinformation all around social media, or impact college admissions. Even if teens aren’t actively thinking about AI, it’s interacting with all these components of their lives.

AI of course can be an exciting thing -- generative AI can help personalize student learning in ways we have never done before: a personal AI tutor bot or customized quiz questions or flashcards for even the most specific of tasks. An essay helper and a creative writing assistant and an in-your-pocket advice giver that's never tired at 4 in the morning for your latest worry about gym class.

But we've also seen an unprecedented amount of AI at our fingertips -- especially on social media platforms which researchers have already been concerned about for over a decade now, specifically its effects on young people. What we need is AI literacy, and fast.

But what is AI literacy, especially for teens? We are both computer science educators, so you might think we think it's very important for kids to learn how neural networks work, pronto! And that's certainly part of it. But AI literacy is much more than knowing how AI algorithms function. Here we have some guidelines on an AI competency framework for students, based on an Understand: Apply : Create framework, along with a human-centered and ethics-based mindset as core competencies.

Druga et al contribute the 4As framework for children and families to critically explore how AI works in their lives – ask, adapt, author, and analyze; with more of a focus on how AI is actually working in their daily life and their family system, as opposed to just ‘here is how AI works’

Research:
AI literacy improves... blah blah blah
Paper: long & magerko 2020
TeachAI stats?
Code. org teaches...

But another thing that students need is a voice in the AI conversation. I always say that we are actually uniquely lucky as educators right now, because our students have so much lived experience with AI technologies. I don't have to teach them using abstract examples I can teach them using something they used yesterday, or this morning. It brings a whole new meaning to hands-on learning.

What students also have is lived experience of AI risks and harms. Those beauty filters that make their skin a little whiter or their eyes a little bluer. Getting accused of using AI on an essay that they wrote themselves. A class bully creating inappropriate photos of them using genAI tools. An AI bot that tells them to cut their calories or work out more than they should, because it doesn't realize they are 13.

Now of course, we can add in Guardrails and Parental Controls. Using a limited data source via RAG is a great technique. Jailbreaking, blacklisting, redteaming, and various alignment techniques can get our AI content safer. There are lots of cool engineery ways to protect kids and we should absolutely be doing them. Microsoft has some excellent resources on Responsible AI and getting started with content safety here [link].

But one thing I know about teenagers, and myself for that matter, is that they don't like being told what to do (CITE). Put a Guardrail or a restriction on their AI, they'll try to break it just because they can. It's fun. Also tools like StealthGPT (sorry kids, we know) that will humanize your writing so you don’t get caught by AI detectors. There’s AI for everything, even AI to seem like it’s not AI.

So instead, we need to tap in to those lived experiences of harm and self-reflection for teens to get the bigger picture. How could this AI system go wrong, and where do I feel empathy for myself and others?

Some research of mine actually explored the kinds of algorithmic harms that show up in your own social media data. Because of GDPR, you can actually download your social media data and take a look at your ad recommendation topics. I got participants from a wide range of backgrounds to take a look at their own ad data and report back on what they found. They did this while simultaneously learning how recommender systems work, how AI decides what ad you're going to see. Participants reflected on all kinds of AI harms: echo chambers and filter bubbles; misinformation and not being able to tell what was real; dieting content that keeps getting more and more specific; privacy risks with their LGBTQ identities; and even 'greenwashing' of products pretending to be more environmentally sustainable than they actually are. But a cool finding was that when you learn a little bit about how the algorithms work, you feel more equipped to change the algorithm in your favor. A win-win for AI literacy and responsible AI.

I also spent some of my PhD research time in the space of social media addiction and its impact on teens. Turns out that I wouldn't quite describe what we are seeing as social media *addiction*, but *attachment*. We are attached to our algorithms just like we are attached to our relationships: sometimes clingy and afraid, and sometimes distant and dismissive. Teens can feel an obsessive or hypervigilant emotional reaction to staying on top of their social media presence -- and need to be taught secure attachment strategies when it comes to algorithmic visibility: "it's not you, it's the algorithm."

I did some work looking at the AI topics students cared about most, and found to be the most relatable to their lives. The key findings of that work in progress were that personal identity really impacts how much you care about a specific AI error; such as female participants caring more about Google translate saying 'he is a doctor, she is a nurse' than the male respondants did.
Students cared about really obvious unfair outcomes, like a hiring tool biased against women or a tool that wrongfully judges incarcerated people. They also cared about the environmental impacts of AI and content moderation getting things wrong. There's much more work to be done there, but our findings indicate that older teens can grapple with these AI harms, but need much more guidance on how to fix them in practice. Our young people care about their futures being just.

And finally, my research focuses on the power of self-advocacy from within AI systems. Teach how the AI system works, teach how it can go right or wrong, and then let the student *express themselves*. Participants learned about faulty algorithmic systems for academic success and financial aid, and wrote letters to explain what was wrong with the algorithm and how the 'engineers' should repair it.

We know that storytelling is an incredibly powerful medium for young learners. Shoutout to two of my good friends from the UW Information School who studied the intersection of technology and storytelling:

Dr. Julia Dunbar, showing how pediatric kidney transplant patients wanted to track their medical journeys with much more than statistics and numbers. They wanted to track their journeys with stories, photos, and a general sense of returning to feeling like themselves. And Yolanda Barton, who does incredible storytelling with VR and AR, telling Seattle Black history through immersive technologies and reconnecting our communities with their own music history.

Dan and I took these powerful ideas of storytelling, self-advocacy, and kids engaging with real-world events; and ran with it for a lesson on generative AI for highschoolers. In the Societal Impacts of Generative AI lesson, kids get 'Character Cards' and an AI scenario. For example, a hospital wants to use an AI chatbot to help its patients. Students roleplay as a Doctor, a Lawyer, a Patient, and the Hospital Director. Another example is a Beauty Filter company using AI to enhance images. Students play as a Highschooler, a CEO, a Science Youtuber, and a Photographer. They all work together to 'regulate' the AI technology like a government or community guidelines would.

So they've learned about AI algorithms and they've learned about AI ethics, but now they get to express their opinions for these different AI scenarios: using their knowledge, experiences, feelings, designs, and even technical solutions to come up with their policies.

I had the privilege of running this activity for University of Washington's Misinformation Day: a big event from UW's Center for an Informed Public that brings together 500 highschoolers to learn about media literacy, misinformation, and AI risks. Students get a day of activities put on by PhD students and faculty, including the Misinformation Escape Room which you can download for free! Last year, students worked together to figure out if a trending new medicine was a hoax or real, analyzing charts and identifying AI-generated images to determine fact or fiction. Next they got to play at my station where they became doctors and lawyers and teachers and politicians and helped regulate AI.

Students have lots to say when it comes to how we can repair AI risks -- allow them to be part of the conversation.

We know that the AI sector is still working on inclusiveness and diversity: while the gender and race gaps are improving, we have a long ways to go. [CHARTS HERE]. According to [LONG 2020, GARRETT 2020, REGISTER 2024, DRUGA 2021 ...] one of the ways we mitigate AI bias is through AI literacy that includes lived experiences.
https://wip.mitpress.mit.edu/pub/the-4as/release/1

Key Takeaways:

- ai risks pop up every day, and we need good engineering to mitigate them
- but guardrails and controls aren't the only answer
- we also need to include teens as experts of their own experiences: how would *they* solve these problems? what do *they* think?
- We know this kind of self-exploration, self-advocacy, and storytelling of lived experiences can improve learning outcomes, specifically with AI (cite cite cite)
- use these resources to get started, link link link
- AI is not the center. Our teens are. And they have quite a lot to say if you know how to ask.